Meta-path Analysis on Spatio-Temporal Graphs for Trajectory Prediction

Analyzing meta-path based features on spatio-temporal graphs combined with a structural RNN structure to predict agent trajectories in autonomous scenes. The proposed model demonstrated the usefulness of these features and achieved a performance boost of at least 32% over all baselines.